phytools Correspondence¶

phytools is the live reference for phylogenetic signal, Mk fitting, continuous-trait utilities, stochastic histories, and ancestral-state summaries. The governed registry currently contains 65 cases across 21 shards. Deterministic estimates and stochastic summaries use different acceptance rules.

Governed Surface¶

Territory	phytools references	Correspondence level
signal	`phylosig` with K and lambda	statistics, fitted lambda and named likelihood quantities
discrete models	`fitMk` with ER, SYM and ARD	rate rows, likelihood and supported state ordering
ancestral states	`fastAnc`, `anc.ML`, `rerootingMethod`	node estimates, uncertainty or state probabilities under comparable models
regression and residuals	`pgls.SEy`, `phyl.resid` with BM and lambda	bounded coefficient, covariance and residual summaries
continuous simulation	`fastBM`, `sim.corrs`	moment, variance, covariance and correlation summaries
stochastic histories	`make.simmap`, `countSimmap`, `describe.simmap`, `densityMap`, `sim.history`	transition-count, dwell-time, state-density and history-summary envelopes
comparative tests	`phylANOVA`	declared group and test summaries under governed fixtures

Function Correspondence Key¶

phytools reference	Bijux registry surface	Observation family
`phylosig(method='K')`	`compute_phylogenetic_signal_test`	statistic, permutation population, p-value and exclusions
`phylosig(method='lambda')`	`estimate_pagels_lambda`	fitted lambda, likelihood/profile quantities and bounds
`fitMk` with ER, SYM, ARD	`fit_discrete_mk_model`	labeled transition rates, likelihood and state order
`fastAnc`, `anc.ML`	`reconstruct_continuous_ancestral_states`	structurally keyed node estimates and uncertainty
`rerootingMethod`	`reconstruct_discrete_ancestral_states`	node/state probabilities under aligned root/model semantics
`pgls.SEy`, `phyl.resid`	`run_pgls`, `summarize_phylogenetic_residuals`	coefficients, covariance/error identity and residual rows
`fastBM`, `sim.corrs`	Brownian and correlated-trait collection simulators	replicate-level moments, variance and covariance summaries
`make.simmap`, `countSimmap`, `describe.simmap`, `densityMap`	stochastic-map simulation, counting and summary surfaces	transitions, dwell time, state density and replication envelope
`sim.history`, `phylANOVA`	discrete-history simulation and phylogenetic ANOVA summaries	history distribution or declared group/test observations

The registry binds each row to specific cases. It does not assert a drop-in API or coverage for unregistered arguments, plotting behavior, or default changes in future phytools versions.

Deterministic And Stochastic Evidence¶

For deterministic cases, the registry can compare named scalars, vectors, rates, node values, or probabilities directly within a tolerance. For simulation and stochastic mapping, exact histories are not expected to match across random-number generators. Those cases compare declared distributional or aggregate summaries across governed replication policies.

pgls.SEy is intentionally narrow. It validates the comparable measurement-error PGLS surface; it is not presented as a general phytools::pgls correspondence claim. Broader coefficient comparisons use the checked APE plus nlme reference where that is the actual source.

What A Passing Stochastic Case Means¶

A stochastic case passes when its registered replication policy completes and every declared aggregate or distribution rule holds. It does not require the R and Bijux histories to contain the same random events. The record therefore retains replicate count, seed policy, summary definition, tolerance or envelope, failed replicates, and observed distributional quantities.

The Stochastic Observation Unit¶

The comparison unit is a declared summary over a declared replicate population, not one simulated tree or history. Each observation retains the case, summary statistic, requested and completed replicates, invalid/excluded replicates, reference and Bijux distributions or aggregates, seed policy, comparison rule, threshold, and outcome.

Changing the replicate count, exclusion policy, burn-in-like selection, or summary statistic changes the observation. It cannot reuse a previous pass even when the function, fixture, and headline mean remain the same.

Do not remove non-finite, degenerate, or failed replicates before computing a headline summary unless the registry declares that exclusion. Selected, executed, passed, failed, and skipped case counts remain separate from within-case replication denominators.

Run The Live Lane¶

RUN_LIVE_PARITY=1 bijux-phylogenetics parity \
  --reference-source phytools-live \
  --phytools-shard phytools-fitmk-model-er \
  --summary-out artifacts/parity/phytools-summary.tsv \
  --observations-out artifacts/parity/phytools-observations.tsv \
  --phytools-failure-root artifacts/parity/phytools-failures \
  --json

Use --phytools-case for an individual fixture. The live lane requires Rscript, phytools, and RUN_LIVE_PARITY=1.

Programmatic Surface¶

Public locator	Purpose
`bijux_phylogenetics.parity.phytools:list_phytools_parity_cases`	enumerate the 65 deterministic and stochastic contracts
`bijux_phylogenetics.parity.phytools:list_phytools_parity_shards`	enumerate 21 operation-owned shards
`bijux_phylogenetics.parity.phytools:run_phytools_parity_cases`	execute selected live cases under the opt-in boundary
`bijux_phylogenetics.parity.phytools:write_phytools_parity_summary_table`	persist case and shard status totals
`bijux_phylogenetics.parity.phytools:write_phytools_parity_observation_table`	persist scalar, row, distribution, and mismatch observations

Discovery defines the eligible case population before random draws occur. A stochastic case cannot be reselected after viewing its summaries.

Preserve Both Denominators¶

A stochastic parity report has two denominators:

selected registry cases, partitioned into executed, passed, failed, and skipped cases;
within each executed case, requested, completed, invalid, excluded, and summarized replicates.

A case can execute yet fail because too few valid replicates remain. It can pass a distribution rule while still retaining individual invalid draws. Do not substitute the replicate pass fraction for the registry case result, or the case pass fraction for the within-case sampling record.

Separate Monte Carlo Error From Correspondence Error¶

A difference between stochastic summaries can arise from finite replication even when both implementations sample the intended distribution. Predeclare a replicate budget, summary estimator, uncertainty estimate, and decision rule that distinguish expected Monte Carlo variation from a material distributional difference.

Use independent random streams within and across implementations. Matching integer seed labels do not imply matching draws across R and Python, and paired event-by-event comparison is not valid unless the case owns a shared random number contract. Repeat the complete batch with independent seeds when the registered rule requires stability; do not keep the most favorable batch.

Observed pattern	Correct response
summaries agree within the predeclared uncertainty rule	retain the pass with both replicate populations and uncertainty estimates
result changes materially across independent batches	report unstable evidence and increase or redesign the registered replication policy
disagreement persists as Monte Carlo error shrinks	retain a correspondence failure and inspect model, state, and summary semantics
invalid draws differ systematically by implementation	compare the failure-generating population before any filtered summary
agreement appears only after an observed-data exclusion	register a new observation; do not rewrite the original case

Increasing replication after seeing a disagreement is diagnostic work, not a retroactive change to the original acceptance rule. Preserve the original batch and identify any larger follow-up comparison separately.

Boundaries That Remain Visible¶

ER and SYM rerooting cases are compared only where root-prior semantics can be aligned; neighboring models remain native review surfaces without a live claim.
Stochastic-history agreement concerns declared summaries, never exact event placement on every branch.
Density maps and other plotting-adjacent results are compared through their underlying numerical ledgers, not pixel identity.
A passed shard validates its registered fixtures and versioned conventions; it does not certify all phytools arguments or future package behavior.
A passed harness case is not a matched Evidence Book verdict; scientific promotion requires a governed claim, provenance, checks, and freshness.